KNOW.me
Conversational AI for survey data analysis
- Client
- KNOWLIMITS Group
- Sector
- Market & Media Research
- Engagement
- Custom agentic AI on proprietary MML dataset
The depth of the data is its strength — and its bottleneck.
KNOWLIMITS operates one of Czechia's most detailed market & lifestyle survey datasets: 8,500 respondents, ~19,000 variables each, ~159 million data points per study.
Every business question, from “who buys premium wine” to “what does our brand's audience watch in the evening” , required a custom analysis by the research team. Internal stakeholders rarely escalated smaller questions, and external clients had to wait for scheduled deliveries.
KNOWLIMITS asked us a direct question: could an AI assistant open this dataset to anyone on the team, without compromising the accuracy their clients pay for?
A dataset you can talk to.
We designed and built KNOW.me — an agentic AI assistant that turns survey analysis into a conversation in Czech. The system works in two stages:
Define a target group
The agent searches the variable space semantically, surfaces the relevant survey questions, and confirms the segment definition with the user in plain language.
Analyze the segment
The user asks anything — media habits, brand affinity, lifestyle, demographics. The agent retrieves the right columns, writes Python, and executes it directly on the dataset.
Every answer is a real, weighted statistic projected to the full Czech 15+ population. No hallucinated numbers. The LLM decides what to compute; pandas performs the computation.
Agentic orchestration
LangGraph + Claude coordinate retrieval, dialog, and execution.
Semantic retrieval
Sentence embeddings over the full ~19,000-variable space.
Hierarchical navigation
Category-tree fallback for broad, less-specific topics.
Verifiable statistics
Python tool execution on the live dataset — every answer is reproducible.
Population weighting
Outputs projected to Czech Republic 15+ (n ≈ 9.05M).
From a queue to a conversation.
A dataset that was previously accessible only through the research team is now a self-service tool for the entire organisation. Anyone at KNOWLIMITS can interrogate ~159M data points conversationally, and the research team is freed to focus on the questions that genuinely need human judgement and methodological depth.
For research firms sitting on rich proprietary datasets, the constraint is rarely the data — it's the queue.
KNOW.me shows that this queue can be replaced by a conversation, without the accuracy compromises usually associated with LLMs on tabular data.